Formatting Fine-Tuning Datasets: How to Correctly Structure CSVs for OpenAI

Fine-tuning allows developers to customize LLM behaviors, tone, and formatting rules. However, preparing datasets for fine-tuning requires strict data formats. OpenAI and other providers require files formatted in JSONL (JSON Lines), where each row is a complete JSON object containing a conversation array.

The Conversational JSONL Structure

Each line in your fine-tuning dataset must match this exact schema structure:

{
  \"messages\": [
    { \"role\": \"system\", \"content\": \"System prompt guidelines.\" },
    { \"role\": \"user\", \"content\": \"User query example.\" },
    { \"role\": \"assistant\", \"content\": \"Desired model response.\" }
  ]
}

Converting Spreadsheets to JSONL

Typically, training data is collected in spreadsheets or databases with columns like instruction, input, and response. To convert this into fine-tuning JSONL, you must parse the spreadsheet, map these columns to their corresponding conversation roles, validate character counts, and output the result in UTF-8 JSONL. Try our client-side CSV/Excel to JSONL Converter to format your fine-tuning datasets securely.